Rethinking how we see risk.
For Aon, I reimagined a data-heavy climate risk tool as a modern, visual experience, with all the complexity underneath kept intact.
Climate models are complicated enough. The interface didn’t need to be.
The original work is internal and proprietary to Aon. The visuals and prototype on this page are recreations.
The short version
1-minute read · full story below
A lot of important information. Not a lot of breathing room.
Climate risk has to say a lot at once: where, which hazard, how bad, by when, why, and which way it’s heading.
The problem wasn’t a lack of data.
It was helping someone understand what mattered without making them decode the interface first.
Before we move forward, let’s go backwards.
To pressure-test the direction, I imagined what the same information might have looked like in an earlier generation of enterprise software: tabs, tables, nested controls, tiny charts and just enough gray chrome to make Windows 95 proud.
NOTE — A speculative, imagined legacy interface. Aon never shipped this. Thankfully, we’ve moved on from Windows 95.
The data wasn’t the problem.
Finding the story inside it was.
Start with the questions, not the pixels.
I organized the interface around the six questions people bring to a site.
Before it became a system, it was a sketch.
The layout settled on a simple pairing: geography on one side, explanation on the other. The map gives context. The panel says what it means.
Make the map
the interface.
Instead of hopping between reports and tables, people move through risk spatially, from the national picture to one site without losing their place.
Color has a job.
Risk data has never been accused of being charming, and it shouldn’t be alarming either. The palette needed enough contrast to show patterns across thousands of cells, and enough restraint to carry labels, charts and two themes.
The same risk language carries through map cells, scores, bars, badges, tooltips and comparison states.
Zoom out for the pattern. Zoom in for the reason.
At the national level, the map reveals broad patterns. Moving into a hazard layer changes the question from ‘Where is risk concentrated?’ to ‘What is driving risk here?’
Risk isn’t a snapshot.
Time became a control instead of a footnote. Rather than digging projections out of a report, people compare a location across 2026, 2031 and 2036.
Get me to the place I care about.
A map is great for discovery. Search is better when you already know where you’re going, so nobody has to pan across a continent to find one building.
And sometimes you just want to ask.
Not everyone wants to read a legend, compare four hazard bars and summarize a trend line in their head. The conversational layer lets them just ask. It explains the map. It doesn’t replace it.
I gave the AI a deliberately narrow job. It answers only from the scores and drivers the map already shows, so anything it says can be checked against what’s on screen. Deciding what it was allowed to say came before deciding how it should sound.
The visualization shows you what’s happening. The conversation helps explain why.
One product. Two very different moods.
Both themes share one risk system. Surfaces and contrast adapt so severe still reads as severe, on paper-white or near-black.
From data-heavy tool to explorable system.
Location, hazard, time, comparison and explanation, all in one view. Someone can get from ‘What am I looking at?’ to ‘Why does this matter?’ without leaving it.
The prototype below is a recreation of the React.js and three.js build. The original stayed with Aon.
Before the build The deck that got sign-off before the prototype was built. PDFDownload ↓The best visualization doesn’t simplify the data. It simplifies getting to the point.
This was less about making climate risk beautiful and more about making it navigable. The best decisions were about what stays visible, what can wait, and how geography, time and explanation work together instead of competing.
That’s the part AI doesn’t do for you. It can produce a hundred views of the same data in minutes. Someone still has to know which one helps a person understand their risk and act on it.
That’s why I like prototyping complex systems. Somewhere between the raw data and the finished interface, something difficult suddenly makes sense.